Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. An image analysis system for identifying objects belonging to a particular object class in a digital image of a biological sample, the system comprising a processor and memory, the memory comprising interpretable instructions which, when executed by the processor, cause the processor to perform a method comprising: analyzing the digital image for automatically or semi-automatically identifying a plurality of objects in the digital image; analyzing the digital image for identifying, for each object of the plurality of objects, a first object feature value of a first object feature of said object, thereby identifying a plurality of first object feature values; analyzing the digital image for computing a first context feature value, the first context feature value being a derivative of the plurality of first object feature values or of a plurality of other object feature values of the plurality of objects in the digital image or being a derivative of a plurality of pixels of the digital image; inputting both the plurality of first object feature values and the first context feature value of said digital image into a first classifier; and executing the first classifier, the first classifier thereby using the plurality of first object feature values and the first context feature value as input for automatically determining, for each object of the plurality of objects, a first likelihood of said object of being a member of the particular object class, thereby determining a plurality of first likelihoods of the plurality of objects of being a member of the particular object class.
The image analysis system is designed to identify objects belonging to a specific class within digital images of biological samples. The system processes digital images to automatically or semi-automatically detect multiple objects. For each detected object, the system analyzes the image to determine a first object feature value, generating a set of these values for all objects. Additionally, the system computes a first context feature value, which is derived from either the object feature values, other object feature values, or pixel data within the image. Both the object feature values and the context feature value are input into a classifier. The classifier then evaluates these inputs to determine, for each object, the likelihood that it belongs to the specified object class, producing a set of likelihood values for all objects. This approach enhances object classification by incorporating both individual object features and broader contextual information from the image. The system is particularly useful in biological imaging, where accurate identification of specific cell types or structures is critical for research and diagnostic applications.
2. The system of claim 1 , wherein the determination of the plurality of first likelihoods comprises using, by the first classifier, the first context feature value for leveling out one or more first object feature value variations caused by factors other than a membership of the plurality of objects to one of a plurality of object classes.
This invention relates to a system for classifying objects using a first classifier that processes object feature values and context feature values. The system addresses the challenge of accurately classifying objects when their feature values are affected by factors unrelated to their true class membership, such as environmental conditions or measurement noise. The first classifier determines a plurality of first likelihoods for each object, representing the probability that the object belongs to one of several predefined object classes. To improve classification accuracy, the first classifier uses a first context feature value to mitigate variations in the object feature values that arise from non-class-related factors. This leveling process helps isolate the true discriminative features of the objects, reducing the impact of external influences on classification performance. The system may also include a second classifier that further refines the classification by incorporating additional context features or alternative feature processing techniques. The overall approach enhances the robustness and reliability of object classification in real-world scenarios where feature variations are common.
3. The system of claim 1 , wherein the method further comprises: a) analyzing the digital image for identifying, for each object of the plurality of objects, a second object feature value of a second object feature of said object, thereby identifying a plurality of second object feature values; b) analyzing the digital image for computing one or more second context feature values, each second context feature value of the one or more second context feature values being a derivative of the plurality of second object feature values or of the plurality of other object feature values of the plurality of objects in the digital image or being a derivative of the plurality of pixels of the digital image; c) inputting both the plurality of second object feature values and the one or more second context feature values of said digital image into a second classifier; d) executing the second classifier, the second classifier thereby using the plurality of second object feature values of and the one or more second context feature values for automatically determining, for each object of the plurality of objects, a second likelihood of said object of being a member of the particular object class, thereby determining a plurality of second likelihoods of the plurality of objects being a member of the particular object class; and e) computing, for each of the objects of the plurality of objects, a combined likelihood of being a member of the object class from the first likelihood and the second likelihood computed for said object, thereby determining a plurality of combined likelihoods of the plurality of objects of being a member of the particular object class.
This invention relates to image analysis systems for classifying objects within digital images. The system addresses the challenge of accurately identifying and categorizing objects in complex scenes by leveraging multiple layers of feature extraction and classification. The system first analyzes a digital image to extract a set of object features for each object detected, such as shape, texture, or color. It then computes context features derived from these object features or from the image pixels themselves, capturing relationships between objects or broader scene context. These features are input into a classifier, which generates a likelihood score for each object belonging to a specific class. The system further refines this process by analyzing the image again to extract additional object features and context features, which are processed by a second classifier to produce a second set of likelihood scores. The final classification for each object is determined by combining the likelihood scores from both classifiers, improving accuracy by integrating multiple sources of information. This approach enhances object detection and classification in scenarios where objects may be partially occluded, vary in appearance, or interact with their surroundings.
4. The system of claim 3 , wherein each object of the plurality of objects has assigned at least one further object feature, the system comprising the first classifier and the second classifier and comprising a further classifier for each of the further properties, wherein the method further comprises: repeating steps a) to d) for each of the further properties for respectively calculating a further likelihood of each object of being a member of the object class, thereby calculating a plurality of further likelihoods of each object of the plurality of objects being a member of the particular object class; and computing, for each of the objects, a combined likelihood of being a member of the object class from the first plurality of likelihoods, the second plurality of likelihoods, and the further plurality of likelihoods.
The invention relates to a classification system for objects, addressing the challenge of accurately determining object membership in a specific class by leveraging multiple classifiers and features. The system processes a plurality of objects, each with at least one primary feature and at least one additional feature. A first classifier evaluates the primary feature to generate a first set of likelihoods indicating the probability of each object belonging to a target class. A second classifier assesses a secondary feature to produce a second set of likelihoods. For each additional feature, a dedicated classifier computes further likelihoods. The system then combines these likelihoods—from the primary, secondary, and additional classifiers—to derive a final, comprehensive likelihood for each object. This approach enhances classification accuracy by integrating multiple feature-based assessments, reducing reliance on any single classifier or feature. The method involves iterative processing for each feature, ensuring all relevant data contributes to the final classification decision. The system is designed to handle complex object classification tasks where multiple attributes influence membership in a class, improving robustness and precision in real-world applications.
5. The system of claim 1 , wherein the digital image is an image of a biological sample or a whole tissue slide; or wherein the digital image is a sub-region within an image of a biological sample or a sub-region within a whole tissue slide.
The invention relates to a digital imaging system for analyzing biological samples or whole tissue slides. The system captures or processes digital images of biological samples, which may include entire tissue slides or specific sub-regions within such slides. The system is designed to enhance the analysis of biological specimens by focusing on either the full sample or targeted areas of interest, improving diagnostic accuracy and efficiency. The digital images are processed to extract relevant biological data, which can be used for medical research, pathology, or clinical diagnostics. The system may include imaging hardware and software components that enable high-resolution imaging and precise analysis of cellular structures or tissue morphology. By capturing detailed images of biological samples or specific regions within them, the system facilitates better visualization and interpretation of biological features, aiding in disease detection, treatment planning, and scientific research. The invention addresses the need for accurate and efficient digital imaging solutions in biological and medical applications.
6. The system of claim 5 , the method comprising: selecting the sub region by automatically or manually identifying a sub region of the digital image including a second plurality of objects having a lower heterogeneity in respect to one or more of their properties than a plurality of other objects within one or more other sub regions of said digital image; and using the identified sub region as the digital image for which the first context feature value, a second context feature value and/or one or more further context feature values are calculated.
This invention relates to image processing systems that analyze digital images to extract context features from specific sub-regions. The problem addressed is the challenge of accurately identifying and processing sub-regions within an image that exhibit lower heterogeneity in object properties, such as color, texture, or shape, compared to other areas. This is useful for tasks like object recognition, segmentation, or feature extraction where homogeneous regions provide more reliable data. The system first processes a digital image containing multiple objects. It then selects a sub-region by either automatic or manual identification, focusing on areas where objects share similar properties (e.g., color, texture, or shape) to a greater degree than in other sub-regions. This selection ensures that the chosen sub-region has lower heterogeneity, making it more suitable for accurate feature extraction. The system then calculates context feature values—such as color histograms, texture descriptors, or spatial relationships—for the identified sub-region. These features can be used for further analysis, such as object classification or image segmentation. The method improves upon prior approaches by dynamically adapting to the image content, ensuring that only the most relevant and homogeneous sub-regions are analyzed. This reduces noise and enhances the accuracy of derived features. The system can be applied in various domains, including medical imaging, surveillance, and autonomous navigation, where precise feature extraction is critical.
7. The system of claim 1 , wherein the first context feature value indicates a first variation in the plurality of first object feature values caused by inter-image variation; and/or wherein a second context feature value indicates a second variation in a plurality of second object feature values caused by inter-image variation; and/or wherein each further context feature value of one or more further context feature values indicates one or more further variations in a plurality of respective further object feature values caused by inter-image variation.
This invention relates to a system for analyzing variations in object features across multiple images, addressing the challenge of distinguishing between meaningful object differences and variations caused by factors like lighting, perspective, or noise. The system extracts context features that represent inter-image variations in object feature values, allowing for more accurate object recognition or comparison. Specifically, the system identifies a first context feature value that quantifies a first variation in a set of first object feature values due to inter-image differences. Similarly, a second context feature value quantifies a second variation in a set of second object feature values caused by inter-image factors. Additional context feature values may further capture variations in other sets of object feature values, enabling the system to account for multiple sources of inconsistency across images. By isolating these variations, the system improves the reliability of object detection, tracking, or classification tasks in applications such as surveillance, medical imaging, or autonomous navigation. The approach ensures that variations in object features are properly attributed to external factors rather than inherent object differences, enhancing the robustness of image analysis.
8. The system of claim 1 , wherein computing the first context feature value comprises computing a statistical average of the first object feature values of the plurality of objects or pixels in the digital image; and/or wherein computing a second context feature value comprises computing a statistical average of a plurality of second object feature values of the plurality of objects or pixels in the digital image; and/or wherein computing each further context feature value of one or more further context feature values comprises computing a statistical average of a respective plurality of further object feature values of the plurality of objects or pixels in the digital image.
The invention relates to image processing systems that analyze digital images by computing context feature values based on statistical averages of object or pixel feature values. The system processes a digital image containing multiple objects or pixels, each with associated feature values. To derive context features, the system calculates statistical averages of these feature values. For example, a first context feature value is computed by averaging first object feature values across the objects or pixels in the image. Similarly, a second context feature value is derived by averaging second object feature values, and additional context features are computed by averaging respective further object feature values. This approach enables the system to extract meaningful contextual information from the image by aggregating individual feature values, which can be used for tasks such as object recognition, image segmentation, or other image analysis applications. The statistical averaging ensures that the context features represent broader patterns or trends within the image, improving the accuracy and robustness of subsequent processing steps.
9. The system of claim 1 , wherein the particular object class is one of: a lymphocyte cell, a tumor cell, a cell of a particular tissue type, a cell positively stained with a particular biomarker, a nucleus of any one of said cell types.
This invention relates to a system for identifying and analyzing specific biological objects, such as cells or cellular components, within a sample. The system addresses the challenge of accurately detecting and classifying different types of cells or cellular structures in biological imaging, which is critical for applications in medical diagnostics, research, and treatment monitoring. The system is designed to process images of biological samples to identify objects belonging to a particular class. The object classes include lymphocyte cells, tumor cells, cells of a specific tissue type, cells positively stained with a particular biomarker, or nuclei of any of these cell types. The system uses image analysis techniques to distinguish these objects from other elements in the sample, enabling precise identification and quantification. By categorizing cells based on their type, staining properties, or structural features, the system supports applications such as cancer detection, tissue analysis, and biomarker-based diagnostics. The ability to differentiate between cell types and their nuclei enhances the accuracy of biological assessments, improving diagnostic reliability and research outcomes. The system's flexibility allows it to adapt to various biological imaging modalities, making it suitable for diverse laboratory and clinical settings.
10. The system of claim 1 , wherein each first object feature value of the plurality of first object feature values corresponds to one of: i. an intensity value of the object, the intensity value correlating with the amount of a stain or a biomarker bound to the object; ii. a diameter of the object; iii. a size of the object; iv. a shape property; v. a distance of an object to a next neighbor object; and vi. a texture property; and wherein in case a plurality of second object feature values and/or a plurality of further object feature values is analyzed, each second object feature value of the plurality of second object feature values or each further object feature value of the plurality of further object feature values corresponds to a remaining one of properties i-vi.
The system is designed for analyzing objects, such as cells or particles, in a biological or chemical context, particularly in applications like microscopy or biomarker detection. The problem addressed is the need to accurately characterize and differentiate objects based on multiple measurable features to improve analysis, such as identifying stained cells or detecting biomarkers. The system processes a plurality of first object feature values, where each value corresponds to a specific property of the object. These properties include intensity values (correlating with stain or biomarker binding), diameter, size, shape, distance to neighboring objects, and texture. If additional sets of feature values (second or further object feature values) are analyzed, each value in these sets corresponds to a different property from the same list. This ensures comprehensive multi-parameter analysis, allowing for precise object classification or quantification. The system enables detailed feature extraction and comparison, enhancing the accuracy of biological or chemical assays.
11. The system of claim 1 , wherein the method further comprises: inputting one or more first, second, and/or further properties into a respective classifier, wherein the first, second and/or further properties are: specified manually; or specified by an advanced feature discovery method; or specified by a minimum redundancy and maximum relevance (mRMR) rules.
This invention relates to a system for classifying data using machine learning, addressing the challenge of selecting relevant input features to improve classification accuracy. The system includes a classifier that processes input properties to categorize data, and these properties can be determined in multiple ways. The first method involves manual specification, where a user directly selects the properties to be analyzed. The second method uses an advanced feature discovery technique, which automatically identifies significant properties from the data. The third method applies minimum redundancy and maximum relevance (mRMR) rules to select properties that maximize relevance to the classification task while minimizing redundancy among them. The system dynamically adapts to different input methods, ensuring flexibility in feature selection. This approach enhances classification performance by optimizing the input properties, whether through user input, automated discovery, or rule-based selection. The invention is particularly useful in applications requiring precise and efficient data classification, such as medical diagnostics, financial analysis, or industrial quality control.
12. The system of claim 11 , wherein the one or more first, second, and/or further object features of the plurality of objects: vary within all objects in the digital image; and/or vary within objects of the same digital image, the digital image being a whole slide image; and/or vary within objects of the same digital image, the digital image being a sub-region of a whole slide image; and/or vary within objects belonging to different digital images derived from different tissue samples of the same organism; and/or vary within different individuals of the same species.
This invention relates to a system for analyzing digital images, particularly whole slide images (WSIs) or sub-regions of WSIs, to identify and characterize objects within the images. The system addresses the challenge of detecting and classifying objects in medical or biological imaging where object features may exhibit significant variability. The system processes digital images to extract one or more features from a plurality of objects, where these features can vary in multiple ways: across all objects in a single image, within objects of the same image (whether a whole slide or a sub-region), between objects in different images derived from the same tissue sample, or even between objects in images from different tissue samples of the same organism. Additionally, the system accounts for variability across different individuals of the same species. The system's ability to handle such diverse feature variations improves the accuracy and reliability of object detection and classification in medical imaging applications, such as pathology or biological research. The system may include preprocessing steps, feature extraction, and analysis modules to ensure robust performance despite the inherent variability in biological samples.
13. The system of claim 1 , the method further comprising generating the first classifier by: reading, by an untrained version of the first classifier, a plurality of digital training images from a storage medium, each training digital image comprising a plurality of pixel blobs respectively representing objects of one or more different object classes, each pixel blob being annotated as a member or as a non-member of the object class; analyzing each of the training digital images for identifying, for each annotated pixel blob, a training first object feature value of the first object feature of said pixel blob; analyzing each of the training digital images for computing a training first context feature value, the training first context feature value being a derivative of the training first object feature values or of other training object feature values of a plurality of pixel blobs in said training digital image or being a derivative of a plurality of pixels of the training image; and training the untrained version of the first classifier by inputting, for each of the pixel blobs, at least the annotation, the training first object feature value and the training first context feature value to the untrained version of the first classifier, thereby creating the first classifier, the first classifier being configured to calculate a higher likelihood for an object of being a member in a particular object class in case the first object feature value of said object is more similar to the training first object feature values of the pixel blobs annotated as being a member of said particular object class than to the training first object feature values of pixel blobs annotated as not being a member of said particular object class, whereby the likelihood further depends on intra-image context information contained in the first or other context feature values.
The invention relates to a machine learning system for object classification in digital images, addressing the challenge of accurately identifying objects within images while considering both individual object features and contextual information from the surrounding image. The system employs a classifier trained using annotated digital images, where each image contains pixel blobs representing objects of different classes. Each pixel blob is labeled as either a member or non-member of its respective object class. During training, the system reads these images and extracts object feature values for each annotated blob. Additionally, it computes context feature values derived from the object features of multiple blobs within the same image or from pixel-level data. The classifier is trained by inputting these feature values along with the annotations, enabling it to assign higher likelihoods to objects belonging to a particular class when their feature values closely match those of annotated members of that class. The likelihood also depends on contextual information derived from the image, improving classification accuracy by leveraging intra-image relationships. This approach enhances object detection by integrating both local and global image context, distinguishing it from systems that rely solely on individual object features.
14. The system of claim 1 , wherein the calculating of the first likelihood of the first classifier comprises using, by the first classifier, the first context feature value for leveling out first object feature value variations caused by factors other than the membership of the object to one of a plurality of object classes; and/or wherein the training of the untrained version of the first classifier comprises identifying, by the first classifier, one of a plurality of properties capable of increasing the classification accuracy of the first classifier using the first object feature for classifying objects, the identified property increasing the classification accuracy by leveling out first object feature value variations caused by factors other than the membership of the object to one of a plurality of object classes, the training comprising modifying a classifier model of the first classifier in a way that an input context feature value calculated for the identified property modulates the likelihood calculated for an object by using the first object feature value of said object as input.
This invention relates to machine learning systems for object classification, specifically addressing the challenge of improving classification accuracy by accounting for variations in object features that are unrelated to the object's class membership. The system includes a classifier that calculates the likelihood of an object belonging to a class by using both object features and context features. The context features help mitigate variations in object feature values caused by factors other than class membership, thereby improving classification accuracy. During training, the classifier identifies properties of the context features that enhance accuracy by reducing irrelevant variations in the object features. The trained classifier then uses these properties to adjust the likelihood calculation, ensuring that context feature values modulate the output likelihood based on the object's feature values. This approach allows the classifier to focus on relevant distinctions between object classes while minimizing the impact of extraneous variations. The system is designed to improve the robustness and reliability of classification tasks in scenarios where object features may be influenced by external factors.
15. An image analysis method for identifying objects belonging to a particular object class in a digital image of a biological sample, the method being performed by a processor of an image analysis system and comprising: analyzing the digital image for automatically or semi-automatically identifying a plurality of objects in the digital image; analyzing the digital image for identifying, for each object of the plurality of objects, a first object feature value of a first object feature of said object, thereby identifying a plurality of first object feature values; analyzing the digital image for computing one or more first context feature values, each first context feature value being a derivative of the plurality of first object feature values or of a plurality of other object feature values of the plurality of objects in the digital image, or being a derivative of a plurality of pixels of the digital image; inputting both the plurality of first object feature values and the one or more first context feature value of said digital image into a first classifier; and executing the first classifier, the first classifier thereby using the plurality of first object feature values and the one or more first context feature values as input for automatically determining, for each object of the plurality of objects, a first likelihood of said object of being a member of the particular object class, thereby determining a plurality of first likelihoods of the plurality of objects of being a member of the particular object class.
The invention relates to image analysis techniques for identifying objects in biological samples, particularly for classifying objects into specific categories. The method addresses challenges in accurately detecting and classifying objects in complex biological images, where traditional feature-based approaches may struggle due to variability in object appearance and context. The method involves analyzing a digital image of a biological sample to automatically or semi-automatically detect multiple objects within the image. For each detected object, a first feature value is computed, representing a specific characteristic of the object. Additionally, context features are derived from the object features or pixel data, capturing relationships between objects or broader image context. Both the object feature values and context features are input into a classifier, which evaluates the likelihood that each object belongs to a predefined class. The classifier outputs a probability for each object, enabling classification based on both individual object traits and contextual information. This approach improves accuracy by leveraging both local and global image information, enhancing the reliability of object classification in biological imaging applications. The method is implemented by a processor in an image analysis system, ensuring automated or semi-automated operation.
16. A non-transitory computer readable storage medium for storing computer-executable instructions that are executed by a processor to perform operations, the operations comprising: analyzing a digital image for automatically or semi-automatically identifying a plurality of objects in the digital image; analyzing the digital image for identifying, for each object of the plurality of objects, a first object feature value of a first object feature of said object, thereby identifying a plurality of first object feature values; analyzing the digital image for computing one or more first context feature values, each first context feature value being a derivative of the plurality of first object feature values or of a plurality of other object feature values of the plurality of objects in the digital image, or being a derivative of a plurality of pixels of the digital image; inputting both the plurality of first object feature values and the one or more first context feature value of said digital image into a first classifier; and executing the first classifier, the first classifier thereby using the plurality of first object feature values and the one or more first context feature values as input for automatically determining, for each object of the plurality of objects, a first likelihood of said object of being a member of a particular object class, thereby determining a plurality of first likelihoods of the plurality of objects of being a member of the particular object class.
This invention relates to computer vision and object recognition in digital images. The problem addressed is the accurate classification of objects within an image by leveraging both individual object features and contextual information derived from the image. The solution involves a multi-step process that begins with analyzing a digital image to identify multiple objects and extract specific feature values for each object. These features may include shape, color, texture, or other distinguishing characteristics. Additionally, the system computes context features, which are derived from the object features or pixel data, to capture relationships between objects or broader image context. Both the object feature values and context feature values are input into a classifier, which then determines the likelihood of each object belonging to a specific class. This approach improves object recognition by incorporating contextual information alongside individual object features, enhancing accuracy in distinguishing between different object classes. The system is implemented via computer-executable instructions stored on a non-transitory medium and executed by a processor.
Unknown
February 18, 2020
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.